Trending Articles
Trending Topics
Trending
Enterprise AI is starving for context. Data alone isn’t enough – not when you’re dealing with sprawling ERP landscapes, manual workarounds and processes that rarely run the way they look on a slide. That’s why Celonis hammered home a blunt message at Celosphere 2025 in Munich: “No AI without PI”.
Over the past year, the company says it has traveled the globe, working with more than 200 customer teams and over 40 partners through 30 AI labs in 17 countries to deploy AI in their platforms, all powered by process intelligence.
The core belief is simple: business processes are the connective tissue that make AI useful, safe and scalable. Or, as Chief Product Officer Daniel Brown put it on stage:
“The enterprise of the future won’t be built by a single vendor but will be powered by an open ecosystem with process intelligence as the connective tissue.”
Celosphere 2025, held on October 23–24, was effectively a live blueprint of that vision in action. Here are 7 takeaways from the session.
Multiple studies now put AI project failure rates somewhere between 70% and 95%, largely because models are bolted onto messy processes and legacy systems without real operational understanding.
That’s the backdrop for Celonis’ core message at Celosphere 2025: “No AI without PI.” Celonis’ view is that AI doesn’t just need more data; it needs to understand how work actually flows across ERP, CRM, supply chain, service and all the human “glue” in between.
That’s the role of the Celonis Process Intelligence Platform and its Process Intelligence Graph – a living digital twin that combines operational data with KPIs, business rules and org context to show how the company really runs. AI agents then plug into this twin instead of guessing from siloed datasets, turning process intelligence into the missing context layer for enterprise AI.
At Celosphere 2025, Celonis pushed hard on the idea that a digital twin can’t just be a fancy event log. The Process Intelligence Graph is evolving into a multimodal digital twin that pulls in far more than ERP transactions. It now connects:
Celonis is also expanding its catalogue of pre-built objects and events across processes such as lead-to-order, transportation management and customer service. The result is a richer, system-agnostic view of operations that AI agents can actually reason over, not just report on.
AI is only as good as the pipes feeding it, and Celonis spent a lot of time on those pipes. Data Core, now generally available, is the high-performance layer that lets Celonis extract from “any source system” and query billions of records at enterprise speed. In the keynote, Barclays was showcased running sub-minute refreshes on critical risk and investigation processes, while BAT scaled from 300GB to 15TB of data – a 50x increase – without losing performance.
The bigger story, though, is zero-copy, bi-directional integration. Celonis now plugs directly into Microsoft Fabric and Databricks without duplicating data, letting customers keep their existing lakes as the source of truth while feeding and receiving process intelligence. That matters for AI teams who want live, governed data – not another shadow warehouse to maintain.
If classic process mining shows you one lane of traffic, object-centric process mining (OCPM) shows you the entire junction. At Celosphere 2025, Celonis introduced new OCPM capabilities – including Performance Spectrum, Instance Explorer and the Object-Centric Performance app – to analyze how orders, shipments, materials, and invoices interact at critical handover points.
This matters because many failures don’t happen inside a single process; they happen where processes intersect: transport to storage, storage to packing, packing to shipping, and so on. Paired with always-on process adherence checks for SLAs, segregation-of-duties violations and routing changes, Celonis is pitching OCPM as a way to decide where AI should be deployed based on real friction, not on whoever shouts loudest in the steering committee.
Celonis wants AI to run the business, not just describe it. That’s where the Orchestration Engine and agentic workflows come in. Once process issues are identified and redesigned, Orchestration Engine executes the new flow end to end – coordinating humans, systems, and AI agents in real time.
The Keystone Steel demo delivered during the keynote made this concrete: when outclassed material is at risk of being scrapped, Celonis automatically triggers AI agents to match the right customer, propose a discounted offer, send it through a next-gen customer portal, and, on acceptance, kick off confirmation, fulfillment, delivery and payment. A control tower monitors scrap, sales and waste, while Celonis continuously mines agent–customer interactions to refine behaviour.
Layered with governance features like application lifecycle management and the new Process Intelligence MCP server feeding live context to agents, this is Celonis’ pitch for safe, composable AI-driven operations.
Celonis is trying to turn AI transformation from a slideware promise into a repeatable loop: analyze, design, operate.
In the analyze phase, process mining and the new object-centric tools answer two questions: “How are my processes really running?” and “Where should AI be deployed?” Always-on adherence checks then flag risks in flight – missed SLAs, SoD violations, routing changes – while there’s still time to act.
Design is where teams model their ideal workflow in Process Designer, with AI assistance suggesting best practices and pinpointing where agents should act versus where humans stay in the loop.
Finally, operate is about execution. Orchestration Engine runs the redesigned process end to end, triggering agents on the right business events and continuously feeding performance data back into the graph. The promise: no more AI pilots stranded from the actual workflow.
The closing act at Celosphere 2025 made one thing clear: Celonis doesn’t want to own the whole AI stack, it wants to wire it together. The new Process Intelligence MCP server was described by Eugenio Cassiano, the Senior Vice President of Strategy and Innovation, as a kind of USB-C for agents, letting any compliant agent plug into Celonis, share context and work together.
On top of that, an expanding ecosystem is building domain-specific solutions: Bloomfilter for governing AI in software delivery, Rollio for resolving process exceptions in tools like Microsoft Teams, Trullion for AI-first lease accounting, Emporix for autonomous commerce, and even customers like American Electric Power composing their own intelligent experiences.
The pattern is consistent: partners bring deep domain expertise; Celonis supplies the horizontal process intelligence. Together, they can not only frame and measure value, but attribute it across AI and composable initiatives. All of which makes “No AI without PI” feel less like a slogan and more like an operating model.
